Re: [Bug-gnubg] "Cannot move, cannot move", reloaded

[Jon Kinsey]

Petr Kadlec wrote:
> I have once again experienced the "cannot move, cannot move" bug, so I
> decided to look at it again. I tried the method I mentioned a few months
> ago, and it still works and deterministically crashes gnubg.
>
> I looked at the stack trace (you can take a look at the attachment if
> the maillist passes it through, but it is not really interesting) and
> the important thing is that there are many (I clicked dices many times)
> identical layers of gnubg interleaved with libgtk. After checking the
> source, IMHO the problem is that during the addition of the move to the
> gamelist, PopMoveRecord is called, which in turn calls GTKSetMoveRecord,
> which contains a message dispatch loop with gtk_main_iteration(). And,
> obviously, during this call another dice click is detected... and here
> we go again. Finally, this seems to end in a crash as PopGame tries to
> delete the move list.
>
> Unfortunately (as I have said before) I never worked with GTK, so I am
> unable to dive into the problem too much. AFAICS, the basic idea of
> calling gtk_main_iteration() deep inside the game logic code seems
> not-really-right to me, but I understand that there is some logic with
> enabling and disabling input, so that it is probably expected to work,
> but ... well, it does not for me.

I finally got it to go wrong!  I'm just about to check in the fix.

You were right on lots of things, it was the gtk loop in setmoverecord.
 This is a great example of gtk 1.3 being rubbish, I had to add the
loop to make sure the list was updated correctly.  So now occasionally
the current move may not scroll in to view.  The way this is called
hints at deeper problems with the structure of the code.

Hopefully we will move on to 2.6 - I'll have a look at the calibrate
dialog/other things soon.

Jon

signature.asc
Description: OpenPGP digital signature